Re: "out of relcache_callback_list slots" after multiple calls to pg_logical_slot_get_binary_changes

Thanks for the comment.

At Wed, 22 Feb 2023 12:29:59 +1100, Peter Smith <smithpb2250(at)gmail(dot)com> wrote in
> On Wed, Feb 22, 2023 at 12:03 PM Kyotaro Horiguchi
> <horikyota(dot)ntt(at)gmail(dot)com> wrote:
> >
> > At Tue, 21 Feb 2023 10:31:29 +0000, "shiy(dot)fnst(at)fujitsu(dot)com" <shiy(dot)fnst(at)fujitsu(dot)com> wrote in
> > > Thanks for your reply. I agree that's expensive. Attach a new patch which adds a
> > > static boolean to avoid duplicate registration.
> >
> > Thank you for the patch. It is exactly what I had in my mind. But now
> > that I've had a chance to mull it over, I came to think it might be
> > better to register the callbacks at one place. I'm thinking we could
> > create a new function called register_callbacks() or something and
> > move all the calls to CacheRegisterSyscacheCallback() into that. What
> > do you think about that refactoring?
> >
> > I guess you could say that that refactoring somewhat weakens the
> > connection or dependency between init_rel_sync_cache and
> > rel_sync_cache_relation_cb, but anyway the callback works even if
> > RelationSyncCache is not around.
> >
>
> If you are going to do that, then won't just copying the
> CacheRegisterSyscacheCallback(PUBLICATIONOID... into function
> init_rel_sync_cache() be effectively the same as doing that?

I'm not sure if it has anything to do with the relation sync cache.
On the other hand, moving all the content of init_rel_sync_cache() up
to pgoutput_startup() doesn't seem like a good idea.. Another option,
as you see, was to separate callback registration code.

> Then almost nothing else to do...e.g. no need for a new extra static
> boolean if static RelationSyncCache is acting as the one-time guard
> anyway.

Unfortunately, RelationSyncCache doesn't work - it is set to NULL at
plugin shutdown.

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center